Reversal of epigenetic aging and immunosenescent trends in humans

Abstract Epigenetic “clocks” can now surpass chronological age in accuracy for estimating biological age. Here, we use four such age estimators to show that epigenetic aging can be reversed in humans. Using a protocol intended to regenerate the thymus, we observed protective immunological changes, improved risk indices for many age‐related diseases, and a mean epigenetic age approximately 1.5 years less than baseline after 1 year of treatment (−2.5‐year change compared to no treatment at the end of the study). The rate of epigenetic aging reversal relative to chronological age accelerated from −1.6 year/year from 0–9 month to −6.5 year/year from 9–12 month. The GrimAge predictor of human morbidity and mortality showed a 2‐year decrease in epigenetic vs. chronological age that persisted six months after discontinuing treatment. This is to our knowledge the first report of an increase, based on an epigenetic age estimator, in predicted human lifespan by means of a currently accessible aging intervention.

For these reasons, we conducted what may be the first human clinical trial designed to reverse aspects of human aging, the TRIIM (Thymus Regeneration, Immunorestoration, and Insulin Mitigation) trial, in 2015-2017. The purpose of the TRIIM trial was to investigate the possibility of using recombinant human growth hormone (rhGH) to prevent or reverse signs of immunosenescence in a population of 51-to 65-year-old putatively healthy men, which represents the age range that just precedes the collapse of the TCR repertoire. rhGH was used based on prior evidence that growth hormone (GH) has thymotrophic and immune reconstituting effects in animals (Kelley et al., 1986) and human HIV patients (Napolitano et al., 2008;Plana et al., 2011). Because GH-induced hyperinsulinemia (Marcus et al., 1990) is undesirable and might affect thymic regeneration and immunological reconstitution, we combined rhGH with both dehydroepiandrosterone (DHEA) and metformin in an attempt to limit the "diabetogenic" effect of GH (Fahy, 2003(Fahy, , 2010Weiss, Villareal, Fontana, Han, & Holloszy, 2011). DHEA has many effects, in both men and women, that oppose deleterious effects of normal aging (Cappola et al., 2009;Forti et al., 2012;Shufelt et al., 2010;Weiss et al., 2011).

| Treatment safety and side effects
A primary concern in this study was whether increased levels of a mitogen (IGF-1) might exacerbate cancerous or precancerous foci in the prostate. Both of these changes should be detectable by measuring PSA or percent free PSA levels. However, PSA, percent free PSA, and the ratio of PSA to percent free PSA, an overall index of prostate cancer risk, improved significantly by day 15 of treatment and remained favorably altered to the end of 12 months (Figure 1ac). A brief spike in PSA at 6 months in two volunteers was rapidly reversed and, after volunteer consultation, was interpreted as reflecting sexual activity close to the time of PSA testing. No change in testosterone levels was observed.
Another significant concern was whether augmenting immune activity might exacerbate age-related inflammation. However, CRP declined with treatment, the decline reaching statistical significance by 9-12 months ( Figure 1d). The pro-inflammatory cytokine, IL-6, did not change (Figure 1e).
No remarkable changes were noted in serum albumin, lipids, hemoglobin, hematocrit, platelet count, electrolytes, and hepatic enzymes ( Figure 1f). Insulin levels were in general adequately controlled by co-administration of DHEA and metformin ( Figure 1g) (although one outlier increased mean insulin at 12 months), and glucose levels did not change ( Figure 1h). Finally, estimated glomerular filtration rates (eGFR), which are relevant to the potential for lactic acidosis with metformin as well as to treatment efficacy, showed a statistically significant improvement after 9-12 months (with a trend toward improvement at 18 months as well) (Figure 1i). Side effects were mild, typical of rhGH administration, and did not require dosing modification except in two cases. Side effects included arthralgias (2 cases), anxiety (1 case), carpal tunnel syndrome (1 case), fluid retention (1 case), mild gynecomastia (1 case), and muscle soreness (1 case). One trial volunteer was removed from the study after approximately one month due to self-reported bradycardia, which preceded the trial, and belated admission of a strong familial history of cancer.

| Thymic and bone marrow regenerative responses
Obvious qualitative improvements in thymic MRI density were observed and are illustrated in Figure 2. Quantitatively, the overall increase in the thymic fat-free fraction (TFFF) was significant at the p = 8.57 × 10 −17 level based on linear mixed-model analysis, implying a restoration of thymic functional mass. The improvements were significant in 7 of 9 volunteers (Figure 3a-c). Two volunteers had abnormally low levels of thymic fat (high TFFF) at baseline, and their TFFFs did not significantly improve with treatment (peak relative changes of +9.6% (p > .3) and +12.4% (p > .2); Figure 3b). Their lack of response was not age-dependent. Instead, improvement in TFFF was dependent upon baseline TFFF, regardless of baseline age (Figure 3c). F I G U R E 1 Treatment safety indices. In this and other figures, error bars depict SEMs; baseline SEMs for normalized data were obtained as described in Experimental Procedures. Asterisks denote p < .05 (*), p < .01 (**), and p ≤ .001 (***). (a) Prostate-specific antigen (PSA). (b) Fold change (FC) in percent free PSA. (c) Fold change in the ratio of PSA to percent free PSA ("risk factor"), which rises as prostate cancer risk rises. (d, e) Pro-inflammatory indices (c-reactive protein (CRP)) and IL-6). (f) Serum alkaline phosphatase (AP), aspartate aminotransferase (AST), and alanine aminotransferase (ALT). The increase in AP, while statistically significant, was quantitatively negligible and remained well within the normal range. (g) Maintenance of insulin levels within the upper and lower limits of the normal range (indicated by the horizontal lines). (h) Lack of change in serum glucose, which remained within the normal range. (i) Improvement in estimated GFR at 9 and 12 months of treatment, with a trend toward continued improvement 6 months after discontinuation of treatment F I G U R E 2 Example of treatment-induced change in thymic MRI appearance. Darkening corresponds to replacement of fat with nonadipose tissue. White lines denote the thymic boundary. Volunteer 2 at 0 (a) and 9 (b) months By comparison, sternal BMFFF increased to a much lesser degree, but with such consistency (Figure 3d) as to reach high statistical significance ( Figure 3e; p < .001 for single-point comparison, or p = 9.5 × 10 −12 for formal linear mixed-model analysis). Bone marrow, similar to thymus, showed a pattern of increased BMFFF with increased baseline fat content, but the details of the pattern were different (Figure 3f). This difference plus the more robust replacement of thymic vs. bone marrow fat is consistent both with a specific reversal of thymic involution rather than generalized regression of body fat owing to GH administration and with possible stimulation of bone marrow T-cell progenitor production by GH (French et al., 2002;Hanley, Napolitano, & McCune, 2005).

| Immune cell subset and cytokine changes
Analysis of CyTOF-defined immune cell populations revealed the most robust changes to be decreases in total and CD38-positive monocytes (Figure 4a, b) and resulting increases in the lymphocyteto-monocyte ratio (LMR) (Figure 4c). Linear mixed-model correlation analysis across all treatment times (0, 9, and 12 months) also showed significant correlations between TFFF and the reduction in total monocyte percentage (r = −0.4, p = 0.039), the elevation in overall LMR (r = 0.45, p = 0.019), the reduction in CD38 + monocyte percentage (r = −0.59, p = 0.0012), and the increased ratio of lymphocytes to CD38 + monocytes (r = 0.51, p = 0.0066). The changes in mean monocyte populations persisted 6 months after discontinuation of treatment (p = .012 for normalized CD38 + monocytes and p = .022 for normalized total monocytes: Table S1), and the increase in LMR remained highly significant at 18 months as well.
PD-1-positive CD8 T cells declined significantly with treatment ( Figure 4d). By gating on naïve T cells, we were able to detect significant increases in both naïve CD4 and naïve CD8 T cells   Note: EA = epigenetic age; A = chronological age; 0 = at zero months (trial onset); 9, 12, 18 = at 9, 12, and 18 months after the trial onset; A 0 = age at trial onset; all results given in years or in years per year. CD31 + CD45RA + CD4 + cells (CD4 recent thymic emigrants, or RTEs) over the course of treatment (linear correlation, Figure S1). We did not detect consistent changes in senescent T cells measured either as CD57 + cells or as CD28cells. However, we did detect a significant increase in serum FGF-21 levels ( Figure S2).

| Epigenetic age regression
Although, on average, trial volunteer epigenetic ages (EAs) were lower than their chronological ages (As) at baseline [(EA-A) 0 < 0, Table 1 The effect of treatment on epigenetic age regression at 12 months was not dependent on age at onset of treatment or on (EA-A) 0 (Table 1) Table 1). Furthermore, comparing the rates of aging regression between 0-9 and 9-12 months showed that, for every age estimator, the rate of aging regression appeared to accelerate substantially with increasing treatment time (Figure 5a-d and year in the first 9 months to −6.48 ± 0.34 years/year in the last 3 months of treatment (p < .005, Figure 5f).

| D ISCUSS I ON
The TRIIM trial was designed to investigate the possibility of thymus regeneration and reversion of immunosenescent trends in healthy aging men while minimizing side effects and any possible risks. Our results support the feasibility of this goal but unexpectedly also bring to light robust evidence that regression of multiple aspects and biomarkers of aging is possible in man. These two observations may be related. restoration of youthful thymic histology (Goff, Roth, Arp, & al., e., 1987;Kelley et al., 1986) and by reversal of age-related immune deficits (Kelley et al., 1986). In humans, the existence of surviving thymic tissue after the age of about 54, which is required for successful thymus regeneration in older individuals, has been questioned (Simanovsky, Hiller, Loubashevsky, & Rozovsky, 2012). Available reports indicating increased thymic CT density and immunological improvements induced by rhGH in HIV patients (Napolitano et al., 2008;Plana et al., 2011), whose thymi are physiologically unusual (McCune et al., 1998), are silent on whether regeneration was observed in individuals over the age of 50. The present study now establishes highly significant evidence of thymic regeneration in normal aging men accompanied by improvements in a variety of disease risk factors and age-related immunological parameters as well as significant correlations between TFFF and favorable changes in monocyte percentages and the LMR, independent of age up to the age of 65 at the onset of treatment. These observations are consistent with the known ability of growth hormone to stimulate hematopoiesis and thymic epithelial cell proliferation (Savino, 2007). Our finding of an increase in FGF-21 levels after 12 months of treatment suggests that thymic regeneration by the present treatment may be mediated in part by this cytokine (Youm, Horvath, Mangelsdorf, Kliewer, & Dixit, 2016), which we believe is a novel finding.
We have not found previous studies associating rhGH administration with an increased LMR or reduced monocyte levels, whereas DHEA administration may actually increase monocyte levels (Khorram, Vu, & Yen, 1997 (Gong et al., 2018), cardiovascular disease (Ji et al., 2017), and stroke (Ren, Liu, Wang, & Gao, 2017)], and are also associated with less generalized inflammation (Gong et al., 2018). Protection from cardiovascular disease has been associated with an LMR above 5 (Gong et al., 2018). In our study, mean volunteer LMRs were below 5 at baseline but were well above 5 at the end of treatment and 6 months after the end of treatment.
Second, the great majority of monocytes are CD38-positive.
CD38 is an NADase ectoenzyme and a degrader of the NAD + precursor, nicotinamide mononucleotide, and increased CD38 expression with age appears to be the primary cause of age-related tissue NAD + depletion in mice and most likely in man as well (Camacho-Pereira et al., 2016). Induction of CD38 with aging has been proposed to be driven by age-related inflammation, and to originate in inflammatory cells residing in tissues (Camacho-Pereira et al., 2016), which in principle might include monocytes. Although many other immune cells express CD38, we did not detect any other CD38 + immune cell population that declined in response to thymus regeneration treatment, suggesting that monocytes may be of particular significance.
Our observation of a decline in CRP in combination with reduced monocyte levels therefore suggests the possibility of an increase in tissue NAD + levels. The mechanism by which CRP was reduced is not clear, but reactivation of negative selection in the thymus could in theory reduce autoimmune-related inflammation; clearance of pro-inflammatory viruses or senescent somatic cells could also be involved. Given the relationship between age-related tissue NAD + depletion and the onset of aging phenotypes (Das et al., 2018;Poljsak & Milisav, 2016), an increased tissue NAD + content could be connected to our observations of reversed epigenetic aging, to the general association of higher LMRs with better health, and to previous observations linking thymus transplantation to regression of various nonimmunological aging processes (Fabris, Mocchegiani, Muzzioli, & Provinciali, 1988). Consistent with this possibility, we observed that strong declines in total and CD38 + monocytes, increases in LMR, and reduced GrimAge all persisted for 6 months following discontinuation of treatment. Positive responses also occurred despite potential complications caused by lymph node aging (Thompson et al., 2019). Therefore, the small increases observed in these cells and in CD4 T-cell RTEs are consistent with the ultimate goal of preventing or reversing the normal age-related collapse of the TCR repertoire at ages just above those of our study population (Naylor et al., 2005).

Treatment-induced declines in
There may be both immunological and non-immunological mechanisms of epigenetic aging reversal. GH, DHEA, and metformin have unique effects that are in opposition to aging, and it is possible that the specific combination of these agents activates a broad enough range of therapeutic pathways to account for the previously unpredictable reversal of epigenetic aging, even independently of the immunological markers we have measured.
In this regard, it must be pointed out that GH and IGF-1 can also have pro-aging effects and that most gerontologists therefore favor reducing rather than increasing the levels of these factors (Longo et al., 2015). However, most past studies of aging and GH/IGF-1 are confounded by the use of mutations that affect the developmental programming of aging, which is not necessarily relevant to nonmutant adults. For example, such mutations in mice alter the normal innervation of the hypothalamus during brain development and prevent hypothalamic inflammation in the adult (Sadagurski et al., 2015) that may program adult body-wide aging in nonmutants (Zhang et al., 2017). It seems unlikely that lowering IGF-1 in normal non-mutant adults can provide the same protective effects. A second problem with past studies is a general failure to uncouple GH/ IGF-1 signaling from lifelong changes in insulin signaling. Human longevity seems more consistently linked to insulin sensitivity than to IGF-1 levels, and the effects of IGF-1 on human longevity are confounded by its inverse proportionality to insulin sensitivity (Vitale, Pellegrino, Vollery, & Hofland, 2019). We therefore believe our approach of increasing GH/IGF-1 for a limited time in the more natural context of elevated DHEA while maximizing insulin sensitivity is justified, particularly in view of the positive role of GH and IGF-1 in immune maintenance, the role of immune maintenance in the retardation of aging (Fabris et al., 1988), and our present results.
Whatever the mechanism of epigenetic age reversal may be, the four selected epigenetic clocks, despite measuring somewhat different features of aging and correlating differently with blood composition and leukocyte telomere length, all showed significant regression of epigenetic age. There was also a marked acceleration of epigenetic aging reversal after 9 months of treatment. The implications of the latter observation remain to be explored. Further, although epigenetic aging reversal appeared to partially regress following discontinuation of treatment according to some epigenetic clocks, this was not true for the GrimAge clock, which best predicts human life expectancy and health span (Lu et al., 2019). It remains to be seen whether follow-up measurements of epigenetic aging using the GrimAge clock will show a persistent 2-year gain in predicted life expectancy or a gradual loss of increased life expectancy compared to baseline. In the latter event, it will be interesting to determine whether repetition or prolongation of the trial treatment might restore or further augment the predicted lifespan gain.
Although epigenetic age does not measure all features of aging and is not synonymous with aging itself, it is the most accurate measure of biological age and age-related disease risk available today.
This justifies the use of epigenetic clocks to estimate the effectiveness of putative aging interventions on a practical timescale. The present study strongly supports this approach, having demonstrated regression of epigenetic age with high statistical significance even in a one-year pilot trial involving only 9 volunteers. However, it will be necessary to verify the present results by replicating them in an appropriately powered follow-up study. Although much more remains to be done, the general prospects for meaningful amelioration of human aging appear to be remarkably promising.

| Volunteer recruitment and screening
Ten nominally healthy adult men from 51-65 years of age were recruited for the study by word of mouth following public announcements of the objectives of the trial. There were two cohorts, the first consisting of seven men and the second comprising three men. Cohort 1 was treated from October of 2015 to October of 2016, and cohort 2 was treated from April 2016 to April 2017.
Volunteers first completed an online screening questionnaire, and qualifying candidates then provided a blood sample for objective confirmation of health eligibility. Candidates passing the second screen attended a meeting to enable physical evaluation, verification of ability to self-inject rhGH, baseline magnetic resonance (MR) imaging of the thymus, two additional blood collections (see Appendix S2), informed consent, and instruction on how to fill out provided medication diaries intended to remind volunteers of dosing schedules and to report any adverse effects or any irregularities in dosing.

| Trial conduct
The TRIIM trial was carried out after approval of the study proto- rhGH (Omnitrope, Sandoz) was provided to trial volunteers and was self-administered 3-4 times per week, depending on side effects, at bedtime along with other study medications. All volunteers were also provided with and asked to take supplements of 3,000 IU vitamin D3 and 50 mg of elemental zinc daily.

| MR imaging and analysis
All imaging scans were performed on the same 3T GE Premier MRI scanner at the Lucas Center for Imaging at Stanford University.
Standardized methods for quantitatively determining thymic fat content have not been previously described, so we employed a computational methodology that has previously been applied to the analysis of bone marrow fat content (Hu, Nayak, & Goran, 2011). These methods for quantifying fat have been shown to be more accurate than standard histopathological assessment by biopsy (Fischer et al., 2012) and provide the thymic fat fraction (TFF) as a number from 0% to 100%. TFF was determined in three replicate central thymic regions at each tested time point and was used to compute the thymic fat-free fraction (TFFF) as 100% -TFF. The same methods were also applied to determine the sternal bone marrow fat-free fraction (BMFFF). For more details, see Appendix S2.

| Determination of epigenetic age
The state of genomic DNA methylation was determined in previ-

| Statistical analysis
We used linear mixed-effects regression to properly account for the longitudinal nature of the information (multiple draws from the same person). When correction for time was not necessary, we used ordinary linear or nonlinear regressions. For comparisons between pretreatment and specific post-treatment endpoints, we employed the t test, the t test for paired comparisons, or alternatives selected by SigmaPlot when the assumptions underlying t tests were not justified. To enable normalized comparisons to month zero baselines, replicate results for each volunteer at time zero were averaged and each replicate result was divided by the mean to produce a population of departures from 100% that could then be used for comparison against later time points. Reported P values do not correct for multiple comparisons because independent hypothesis testing was not carried out in most cases, and in the case of both multiple tests for epigenetic age and significant differences in CyTOF results, all significant test results were highly correlated, making Bonferroni correction overly conservative.

ACK N OWLED G M ENTS
We are grateful for the multifaceted support of Garry Nolan and Sean Bendall. We much appreciate the helpful advice and support provided by Yael Rosenberg-Hasson of the HIMC and the cheerful cooperation of many other individuals at the HIMC, the Lucas Center for Imaging, the SBC, and other institutions on the Stanford campus that enabled volunteer processing and on-campus seminars.
We also thank the reviewers of this paper, who helped us improve the manuscript in important ways. We are particularly thankful for the generous personal donations provided by many individuals and for a supplemental grant from the Life Extension Foundation, all of which enabled additional measurements. The study was supported by Intervene Immune, Inc.

CO N FLI C T O F I NTE R E S T
GMF, RTB, JPW, and SH are shareholders in or have options to purchase shares in Intervene Immune, Inc. GMF and RTB are officers of Intervene Immune and are named in a related Intervene Immune patent application. All other authors declare no competing interests.

AUTH O R CO NTR I B UTI O N S
GMF designed and directed the study, compiled the data for all graphics, prepared the graphics, and wrote the manuscript. RTB managed the study; contributed to its design; obtained digital data for all CyTOF, TFFF, and BMFFF analyses; and contributed to the manuscript. JPW served as the study physician, providing medical oversight, support, advice, and volunteer evaluation, and assisted with trial events. ZG designed and arranged for the antibody panel for CyTOF analysis, directed the CyTOF analyses, and assisted with trial events. SSV designed the TFFF and BMFFF measurement techniques. HM and ML provided guidance concerning correct processing and interpretation of CyTOF data and reviewed and approved our CyTOF methodology.
DTSL and MSK performed all epigenetic measurements and provided helpful insights for the manuscript. SH carried out all epigenetic clock calculations, performed all linear mixed-model correlations, and contributed to the manuscript. All authors approved the manuscript.